TW317631B - Speech encoding device and speech decoding device - Google Patents

Abstract

A speech encoding device for splitting voice input into sound signal data and frequency spectrum data to encode the sound signal data by unit of voice frame, characterized in comprising: an object voice generator for generating an object voice vector from voice input at a vector length responded to a delay parameter; a self-adapting voice encoding directory for generating an self-adapting sound vector at a vector length responded to the above delay parameter from a sound signal generated previously; a self-adapting sound indexer for evaluating a relative disagreement of the object voice vector of a synthesized voice vector obtained from the self-adapting sound vector and indexing out a self-adapting sound vector with a minimum disagreement; and a voice frame sound generator for generating a sound signal at a frame length with the self-adapting sound vector with a minimum disagreement.

Description

317631 A7 B7 V. Description of the invention (1) [Technical Field of the Invention] The present invention is a speech encoding device and a speech tweezers decoding device that compressively encodes a speech signal into a digital signal. [Known Techniques] 9 shows an example of the overall configuration of a voice coding and decoding device that rotates the conventional speech into stage I into spectrum envelope information and sound source signal information, and performs tweezing of the sound source signal information in circle units. It is the same as that shown in Japanese Patent Laid-Open No. 64-40899. I _____ Γ ____- printed by Beigong Consumer Cooperative of the Central Bureau of Economic Affairs of the Ministry of Economic Affairs (please read the precautions on the back before writing this page). In the summary, 1 is the encoder, 2 is the decoder, and 3 is multitasking 4 is the separate structure, 5 is the voice input, 6 is the Fu input medium and 7 is the voice output. The braid device 1 is composed of 8 ~ 15 under Μ. 8 is the embedded pre-supported number analysis analysis structure, 9 is the pre-supported parameter setting mechanism, 10 is the self-supporting audio source codebook, 11 is the self-supporting audio source retrieval structure, and 12 is the error information generation structure. Structure, 13 is the drive sound irrigation script, 14 is to retrieve the sound source search structure, and 15 is the sound signal generation mechanism. In addition, the decoupling device 2 is composed of 16 to 22 under Μ. 16 is the media prediction parameter decoding structure, 17 is the self-contained audio source code book, 18 is the self-contained audio source decoder, 19 is the drive sound source tweezers thin, 20 is the drive sound source decoding broadcast structure. 21 is the sound sea Signal generation structure, 22 is a synthesis filter. Under Μ, the action of the speech code decoding device that divides the above-mentioned known cogwheel into the spectrum envelope information and the sound source signal information, and the sound frame signal information is edited by the rim unit is explained. First, in the tweezers 1, for example, the digital speech signal sampled at 8 kHz is used as the speech round 5 to enter. Sensitive prediction parameter analysis suspected that 8 * analyzes voice_in 5, and extracts the frequency as voice_enveloping the pre-prepared paper size of the paper. The standard of China National Standard (CNS) A4 (210 X 297 mm ) -4 317631 A7 B7_ V. Description of the invention (2) Lan parameters. Next, the linear prediction parameter encoding mechanism 9 quantizes the aforementioned linear prediction parameters, and outputs the corresponding code to the multiplexer 3 * At the same time, Dong Zihua ’s media prediction parameters are output to the self-recorded sound source retrieval structure. 11. Error signal generation mechanism 12, driving sound source retrieval mechanism 14. Secondly, add K description to the compilation of the audio signal information. On the codebook 10 of the self-sounding sea, there is a bet on the sound source signal generated in the past from which the sound structure 15 generates the broadcast structure 15 and will correspond to the input from the sound structure retrieval from the sound structure 11 The self-propagating sound source vector of the length of the circle of the delay parameter 丨 is output to the self-adapting sound source ado cable mechanism 11. Here, if the aforementioned self-entangled sound source cuts out the source signal of the length of the HI frame from the sanpU earlier to the delay #number 1 from 1 to the ft system, when the delay #number 1 is shorter than the length of the nursery frame It is generated by repeating the 1-sample chapter source signal until the _ frame length. _10 (a) is an example of self-supporting sound source vector in case of frame length, and B 10 (b) is 1 <case of self-epileptic response sound source in case of circle frame length to Dong ’s case ” Bureau-consumer-consumer-printed self-recording audio source search agency 11, which is the self-polishing of the parameter 1 of the drawing of 20 彡 1forget 128, for example, which will be embedded by the aforementioned adaptive audio codebook 10 The sound source vector uses the quantized linear prediction parameters input by the aforementioned linear pre-holding parameter tweezers 9 to perform castellation prediction synthesis M to generate speech synthesis to fi. Next, the speech weighting vectors and the aforementioned speech synthesis vectors of speech input vectors cut out by speech_entry 5 for each frame are obtained. Secondly * comparatively evaluate the aforementioned loss of life * to find the delay parameter L and the corresponding gain of the self-walking sound source corresponding to the minimum of the aforementioned disappointment, and the above-mentioned delay parameter L and the gain point of the gain point of the self-speaking sound source _ out On the multiplexer 3, simultaneously generate the self-supporting sound source parameter corresponding to the aforementioned extended parameter L multiplied by the aforementioned self-supporting sound source 堉 5 (please first read the "note" item on the back and then fill in the purchase) This paper size is applicable to China National Standards (CNS) Α4 specification (210Χ297 mm) 7 Β 3 / V Explain clearly that the number-signaling source 15 source tone generator should be the appropriate source of the self-numbered source point and the tone structure. The machine has a letter error. Before and after the input and output, the machine code number is compiled and the number of the reference source is measured. The pre-response adaptability line is described by the user who will make the source code of the 12th letter form the source machine into a suitable number. The number of advancements of the pre-test line is measured by the input and output of the quantifier of the pre-test line 9. The input and output of each difference of the error is 5. Pick up harmony. Momentum and momentum are driven into the output and output into the chorus input language M, the output is quantitative. The upper frame is based on the letter S3 (please read the precautions on the back before filling in this page)-installed. Printed on the driving source codebook 13 by the quasi-authorized Peking Consumer Cooperative, for example, storing N driving source directions 5 generated by random noise, and outputting the corresponding driving source code i input by the driving source retrieval mechanism 14 'S drive sound source to 1. The driving sound source retrieval mechanism 14 is to suspend the N driving sound sources, and the driving sound source input from the driving sound source codebook 13 is quantized using the quantized linear prediction parameters input from the linear prediction parameter stimulating mechanism 9 Linear predictive synthesis K generates a speech synthesis vector. Next, the auditory weighted distortion of the error signal direction 1 and the speech synthesis direction fi input by the error signal generating means 12 are obtained. Secondly, compare and evaluate the aforementioned distortion to obtain the driving source code I and the corresponding driving source gain 7 corresponding to the minimum distortion, and output the encoding of the driving source code I and the driving source gain 7 to the multiplexer 3 At the same time, the driving sound source signal corresponding to the driving sound source code I of the driving sound source code I multiplied by the aforementioned driving sound source gain 7 is generated and output to the sound source signal generating mechanism 15. The sound source signal generation mechanism 15 adds the adaptive sound source signal input by the aforementioned adaptive sound source retrieval mechanism 11 and the aforementioned drive sound source retrieval mechanism 14. The paper size of the book is applicable to the Chinese national standard (CNS &gt; Λ4 specification (210Χ297 ) 6 317631 A7 B7 5. Description of the invention (4) The input driving sound source signal M generates a sound source signal and outputs it to the adaptive sound source code book 10. (Please read the precautions on the back before filling in this page) After the encoding is completed, the multiplexer 3 sends the encoding corresponding to the fi # -ized linear prediction parameter, and the encoding corresponding to the delay parameter L, the driving sound source code I, and the sound source gain point, 7 to the transmission line 6. Next , The operation of the decoder 2 is added with K. First, the separation mechanism 4 that receives the output of the multiplexer 3 outputs the encoding of the linear prediction parameters to the medium prediction parameter decoding mechanism 16; the delay parameter L, the sound source The output of the gain code is output to the adaptive sound source decoding structure 18; and the code of the driving sound source code I and the sound source gain 7 is output to the driving sound source decoding mechanism 20. In the Ministry of Economy Printed linear prediction parameter decoding mechanism 16 printed by the Central Bureau of Industry and Fisheries Cooperative Cooperative, decodes the linear prediction parameters corresponding to the aforementioned linear prediction parameters, and outputs to the synthesis filter 22. The adaptive audio source decoding mechanism 18 is derived from The audio source codebook 17 reads out the adaptive sound source vector corresponding to the aforementioned delay parameter L, and decodes the adaptive sound source gain point from the encoding of the aforementioned adaptive sound source gain / 3, K generates the aforementioned self-dependent audio source vector times K An adaptive sound source signal adapted to the sound source gain / 3 and output to the sound source signal generating mechanism 21. The driving sound source decoding mechanism 20 is a driving sound source vector corresponding to the aforementioned driving sound source code I from the driving sound source code book 19, and from In the encoding of the driving sound source gain 7, the driving sound source gain 7 is decoded, and K generates the driving sound source signal multiplied by the driving sound source vector 7 and output to the sound source signal generating mechanism 21. The sound source signal generating mechanism 21 is added According to the paper standard of the aforementioned self-entangled sound source retrieval agency, the Chinese National Standard (CNS) Α4 specification (210X297 (5) Description of the invention (5) Α7 Β7 20 should be constructed to suit. From the code to the solution of the source wheel tone and move simultaneously, the source of the drive letter before the letter is generated by the tone and the number of the letter K. The source number of the letter should be sourced automatically. Of the chased into the wheel _ 18

The Ministry of Economic Affairs of the People's Republic of China has printed the sound source tweezers codebook 17 and synthesis filter 22 on the Pui Pongong Consumer Cooperative. The synthesis filter 22 synthesizes the sound source signal rounded by the aforementioned sound source signal generating mechanism 21 using the linear prediction parameters input by the linear prediction parameter decoding structure 16 and outputs a speech output 7. In addition, as an improved prior art of the above-mentioned conventional speech encoding and decoding device, and a speech encoding and decoding device that can obtain higher-quality speech output, there are P. Kroon and BSAtal 〃 Pitch predictors with high temporal resolution ^ ilCASSP &quot; 90, pp661-664, 1 990). The conventional speech encoding and decoding device improved by the conventional knowledge is in the structure of the conventional speech encoding and decoding device shown in Fig. 9 * The delay parameter of the retrieval object in the retrieval structure 11 as an adaptive audio source is other than an integer value Non-integer rational numbers are also taken. The adaptive audio source codebooks 10 and 17 interpolate and generate self-appropriating audio source vectors corresponding to the aforementioned delay parameters of non-integer rational numbers between samples of audio signal generated in the past, and the output . Pu 11 shows an example of self-entangled sound source to Dong when the extension parameter 1 is a non-integer rational number. The circle 11 (a) is an example of the case of 12 squat frame length, and the leg 11 (b) is an example of the case of 1 <squat frame length. With this structure, Μ is higher than the accuracy of the voice-in and out-of-pounding period of the voice_in determining the delay parameter, and the self-entangled sound source can be generated, as disclosed in Japanese Patent Laid-Open No. 6 4-40 8 9 9 The installed Dongtong can generate higher-quality voice output 0 (please read the note Ϋ on the back first and then fill in this page) -installed--The line book is again suitable for China National Falcon (CNS) Α4 specification (210X297 Mm) 317631 A7 Αν B7 5. Description of the invention (6) In addition, as other conventional technologies of the conventional speech codec device, there is Japanese Patent Laid-Open No. 4-344699. Circle 12 is a configuration diagram showing an example of the overall configuration of a conventional speech encoding device. In the circle 12, the same parts as those of the circle 9 are denoted by the same symbols, and their explanations are omitted. In Fig. 12, 23 and 24 are drive sound source codebooks, which are different from the drive sound source code book of Fig. 9. Here, the operation of the codec apparatus configured as described above will be described by adding K. Printed by the Beigong Consumer Cooperative of the Central Bureau of Economic Development of the Ministry of Economic Affairs. First, in the encoder 1, the adaptive sound source search mechanism 11, such as the delay parameter 1 in the range of 128 for 20S1, will be set by the aforementioned adaptive sound source codebook 10. The input adaptive sound source uses a linear prediction parameter subdivided by the media prediction parameter encoding mechanism 9 to perform linear prediction synthesis to generate a speech synthesis target. Next, the weighted distortion of the speech input vector and each of the speech synthesis vectors cut out by the speech input 5 in each frame is obtained. Secondly, the aforementioned distortion is compared and evaluated, and the delay parameter L at which the aforementioned distortion becomes minimum and the corresponding adaptive sound source gain / 3 are obtained, and the encoding of the aforementioned delay parameter L and the adaptive sound source gain / 3 is output to the multiplexer 3 And drive the audio source codebook 23, at the same time generate the adaptive sound source signal corresponding to the aforementioned delay parameter L by the adaptive sound source parameter multiplied by the aforementioned adaptive sound source gain / 9, and output to the error signal generation structure 12 and the sound source signal generation mechanism 15 on. On the driving sound source codebook 23, for example, N driving sound source vectors generated by random noise are stored, and the driving sound source input by the driving sound source retrieval mechanism 14 is repeated for each cycle of the aforementioned delay parameter L corresponding to the aforementioned delay parameter L Encode the driving sound source vector of i and output it periodically. Figure 13 (a) shows the case of 9 (please read the precautions on the back before filling in this page) This paper size is applicable to China National Standard (CNS) A4 specification (210X297mm) 317631 Α7 Β7 V. Description of invention (7) Example ^ ^ ^ Amount} Periodic direction (b The sound graph of the source 13 will move in parallel, ， The situation is that the number of samples in the period of the emotional situation is quantified and the whole source is non-sounding, driving into the The delay in the parameters shown in the biometrics is driven by the volume source sound before the electricity is used. It will be transferred to the 4 input 1 ® Institute I 3 machine 2 cable. The code check source, sound source, dynamic sound drive, drive parameter test predictive line, ib component, input wheel, 9, structure, code, parameter test, predictive line, utterance, interrogation, interrogation, 〇Quantity to the number of signs into the letter and the difference between the formation of the input and output of the error into the 12 pre-formed test machine pre-formed line number line letter advance (please read the precautions on the back before filling this page) Economy The weighted distortion of the auditory weight printed vectors of the Employee Consumer Cooperatives of the Ministry of Central Accreditation Bureau. , Compare and evaluate the aforementioned distortion, K find the driving source code I and the corresponding driving source gain 7 I corresponding to the minimum distortion and output the driving source code I and the driving source gain 7 codes to the multiplexer 3, Simultaneously generate the driving sound source signal corresponding to the periodic driving sound source vector of the aforementioned driving sound source code I multiplied by the aforementioned driving sound source gain 7 and output it to the sound source signal generating mechanism 15. After the encoding is completed, the multiplexer 3 will correspond The coding of the aforementioned quantized linear prediction parameters, and the coding corresponding to the delay parameter L, the driving sound source code I, and the sound source gain / 3, 7 are sent to the transmission line 6. Next, the operation of the decoder 2 will be described. First accept The separation mechanism 4 of the output of the multiplexer 3 outputs the coding of the linear prediction parameter to the linear prediction parameter decoding mechanism 16; the output of the delay parameter L and the audio source gain / 3 to the adaptive audio source decoding mechanism 18, Drive the audio source codebook 24; and output the code of the drive source code I and the source gain 7 to the drive source decoding mechanism 20. drive source code 24. There is a drive source code book on the coding side. This paper size is applicable to the Chinese National Standard (CNS) Α4 specification (210 × 297 mm). 10 317631 A7 ___B7 V. Invention description (8) The same N drive source vectors, And each pass corresponding to the aforementioned delay parameter l repeats correspondingly to the driving sound source coded by the driving sound source code I input by the driving sound source decoding mechanism 20 and is cycled, and is output to the driving sound source decoder_20 . The drive sound source decoding mechanism 20 decodes the drive sound source gain from the code of the drive sound source gain 7 to generate the periodic drive sound source vector multiplied by the sound source gain 7 keyed in by the drive sound source book 2 4 The sound source signal is driven and output to the sound source signal generating mechanism 21. The sound source signal generating mechanism 21 adds the self-adaptive sound source signal input by the self-supporting sound source decoding structure and the driving sound source signal M input by the driving sound source decoding structure 20 to generate a sound source signal and output it to Zijin The audio source should be thin on the code 17 and synthesized on the slope device 22. The synthesis filter 22 performs linear prediction synthesis on the sound source signal input by the aforementioned sound source signal generating mechanism 21 using the linear prediction parameter input by the linear prediction parameter decoding mechanism 16, and outputs a voice output 7. [Problems to be Solved by the Invention] The Employee Consumer Cooperative of the Central Ministry of Economic Affairs of the Ministry of Economic Affairs is printed on the above-mentioned conventional voice encoding and decoding device. When searching for the sound source of the Yong code, it can be adapted to the sound source according to the delay parameter. The vector or the driving sound source vector is generated as a circle-length sound source vector at the time of generation. It is subjected to media prediction synthesis to generate a speech synthesis direction Dong * to obtain the speech input direction a and the speech synthesis direction a in the frame length interval Distortion. However, due to the large amount of calculations added to the linear prediction synthesis, there is a box in the sound source search that requires a large amount of calculations. The present invention was created to solve such problems. Its purpose is to obtain 11 (please read the precautions on the back before filling in this X). This paper standard adopts the Chinese national rubbing rate (CNS> A4 specification (210X297 ) 317631 A7 B7 V. Description of the invention (9) A speech encoding device and speech encoding and decoding device that can avoid the deterioration of the quality of speech synthesis when encoding speech and can generate good quality speech synthesis with a small amount of calculation Installer. [Means to solve the problem] To solve the above problem * The speech encoding device of the present invention is provided with: a target speech generating mechanism * a person who generates a target speech vector corresponding to a vector length of a delay parameter from speech input; adaptive Audio source codebook, which generates adaptive sound source vectors of the negotiated length corresponding to the aforementioned delay parameters from the sound source signals generated in the past; from the i-source sound source retrieval mechanism, it evaluates and synthesizes the speech synthesis vectors obtained from the aforementioned adaptive sound source vectors The relative distortion of the aforementioned target stack of speech vectors, and retrieve the adaptive sound source vector with the least distortion; and the frame sound The source generation mechanism is a person who generates a frame length audio signal from the aforementioned self-tuning audio source vector with the smallest distortion. Printed by the Beige Consumer Cooperative of the Central Sample Bureau of the Ministry of Economic Affairs (please read the note ^^ on the back and fill in (This page) In addition, the speech encoding device of the present invention is further provided with: a second target speech generating mechanism that generates a second target speech vector from the target speech to the self-responsive sound source vector whose distortion is minimized; The driving sound source codebook * is used to generate the driving sound source vector corresponding to the vector length of the delay parameter; the driving sound source retrieval structure is to use M to evaluate and the second speech synthesis direction S obtained from the aforementioned driving sound source to 1. The relative distortion of the two target speech vectors, and the search distortion becomes the smallest driving sound source to one; and the second frame sound source generation structure is used to generate the second frame length from the aforementioned driving sound source vector whose distortion becomes the minimum The source of the signal. The voice coding device of the present invention is equipped with: a structure for generating a target voice, which generates a corresponding delay parameter from the voice input. The standard paper size is applicable to the Chinese National Standard (CNS) Α4 specification (210X297 mm) -12 _ _B7_ V. Description of invention (10) The phonetic transcription is directed to one; it drives the audio source codebook, which is used to generate the delay parameter The number of vectors of the driving sound source vector; the driving sound source retrieval mechanism, which uses M to evaluate the distortion relative to the aforementioned target speech vector of the speech synthesis vector obtained from the aforementioned driving sound source to 1, and the retrieval distortion becomes the smallest driving sound source Vectors; and the circle frame sound source generating mechanism, which is from the aforementioned driving sound source with the least distortion to the sound source signal used to generate the circle frame length in JI. In addition, the voice editing device of the present invention further corresponds to the voice input The pitch period determines the length of the target speech vector and the driving sound source vector. In addition, the speech coding apparatus of the present invention further makes the vector length corresponding to the embedding parameter rational. In addition, the voice coding device of the present invention is further configured to generate the target speech. M corresponds to the speech input of the delay parameter for each direction of the ft length. The speech input of the frame is divided, and the average speech input of each direction is added to generate the destination. The voice vector. Printed by the Beigong Consumer Cooperative of the Central Bureau of Economic Development of the Ministry of Economic Affairs (please read the precautions on the back before filling in this page) and the voice tweezers device of the present invention is a further mechanism for generating your voice. The length division corresponds to the speech input that is an integer multiple of the length of the delay parameter, and the speech input K that averages the magnitude of each vector is added to generate the speech vector. In addition, the speech coding device of the present invention further makes the length of the integer multiple of the length of the delay parameter for the delay parameter above the frame length M. In addition, the speech encoding device of the present invention further enables the target speech generating mechanism to determine the average speech input of each vector length according to the special amount of speech input of each length S length with corresponding delay parameters * Μ In order to generate _; _ This paper scale is applicable to the national standard (CNS) Α4 specifications (210Χ297 mm) _ 13 · A7 Printed by the Ministry of Economic Affairs Central Bureau of Standards and Engineering Zhengzheng Consumer Cooperative B7. Invention description (II) 1 S stack The weighting factor 0 * · 1 of the most recent speech | In addition, the speech m-code device 9 of the present invention is more advanced-the feature of speech rounding for each vector length of the delay parameter corresponding to the delay 1 1 1 delay parameter 1 If you have at least voice, please input the power information 1 1 first. Read the I and f. The voice m-code device 9 of the present invention is more advanced-step 1 corresponding to the extension of the extended surface Note 1 * I delay parameter of each-the length of the vector Special manager of voice input contains at least voice intention 1 Item I; Involved phase «Informer 0 Refill 1 and 9 The voice m-code device of the present invention is to further enable 100-standard voice to write a page-mounted 1% mechanism for each length of voice input according to the corresponding delay parameter. At the time 1 1 颚 系 Μ decided to add the average speech input of each vector length to generate a huge 1 I standard speech to the broadcaster at the time of 0 1: The speech encoding device 9 of the present invention is further improved-step by step One hundred phonic speech generators 1 1 The mechanism adds the average speech to each vector length of the delay parameter in Μ • 1 1 Fine-tune the speech of each vector length when inputting _ the time of entering the system 〇1 1 and f The voice m-code device 9 of the present invention further makes the ring frame sound line generating mechanism 9 Μ The aforementioned length of each vector repeats the corresponding delay parameter to 1 I Regularization of the sound source signal of the degree 9 Μ The sound source signal that generates the length of the round frame 0 1 I and t The speech coding device of the present invention further generates the sound source of the circle frame 1 1 into the mechanism 9 Interpolates the corresponding extension parameters between the circle frames The sound source signal of the vector length of No. 1 1 Η The person who generated the sound source signal 0 1 and 9 The device for speech sound of the present invention »It is more advanced — to make i respond to the sound source 1 1 Retrieval mechanism with instrument synthesis synthesizer, use the synthesis The impulse response of the filter 1 1 I • &gt; λ Iteratively calculates the disappointed person ΰ 1 relative to the speech synthesis obtained from the aforementioned white adaptive sound source to Dong Zhong 1 1 to Dong ’s aforementioned B-labeled speech vector 1 This paper scale is applicable to China National Standard (CNS &gt; A4 said grid (210X297mm) V. Description of the invention (12) In addition, the voice arbor brick device of the present invention * is further equipped with »up-sanpling voice input voice_input increase Sampling structure, and goal Generated mesh repeater Mei Mi from speech to be increased by the amount of speech samples in the key. In addition, the voice editing device of the present invention * further includes a sound source signal oversampling mechanism for the upsampling signal generated by the upsampling encounter, and the self-sounding upsampling codebook is generated from the past that has been upsampled The sound source signal generated from the audio frequency vector is also supported. The voice mixing device of the present invention further increases the corresponding sampling parameters of the sampling mechanism to change the sampling magnification. In addition, the speech sound device * of the present invention further makes the Poja sampling structure only change the speech input or the increased sampling rate of the sound source signal within the range of the vector length corresponding to the delay parameter. Printed by the Beigong Consumer Cooperative of the Central Bureau of Standards of the Ministry of Economic Affairs. The tweet code decoding device of the present invention is provided in both the editor and the target speech generation structure, which generates the longest corresponding delay parameter from the voice input. The degree of the target voice is one; the self-carburized audio source compilation book is the one that generates the self-sounding sound source vector corresponding to the aforementioned delay parameter ft length from the sound source signal generated on the occasion; the self-speaking sound source retrieval agency , Is to evaluate the distortions that are most relative to the aforementioned target speech speech synthesis direction δ derived from the aforementioned self-responding sound source to ft, and to retrieve the self-grinding sound source with the least distortion to ft; Constructed from the aforementioned self-responsive sound source vectors with the least distortion, generating a source signal with a circle frame fear degree. On the other hand, it has a decoding side. • An adaptive sound source codebook is used to generate the vector length corresponding to the delay parameter. The source of the self-adaptation of the sound source: the frame sound source is generated from the self-supporting 15 (please read the note ^^ on the back before filling in the purchase) This paper standard is applicable to the Chinese National Standard (CNS) Α4 specification (210Χ297 Mm> 317 631 A7 B7 V. Description of the invention (13) The dare sound source vector printed by the Employee Consumer Cooperative of the Central Central Bureau of Economic Affairs of the Ministry of Economic Affairs. Originally, the second sound source vector was the first distortion obtained by driving the sound source with K students, while the frame sound source used M The source side has a directional 1 length that generates a vector from the drive and the original speech; the source vector is the distortion of the sound source to Dong, and the sound source is generated to generate the target speech of the K invention in the drive source vector with the drive degree In the second generation, the search mechanism of the second-synthesis-corresponding search distortion distortion mechanism is generated in the second graph, the driving mechanism of the driving sound source vector invention is the driving sound source; the driving sound source is obtained and the retrieval mechanism is lost, the frame The length of the sound source coded sound source vector is generated by M as a frame length sound source signal decoder device, which is a further generation mechanism that is derived from the target speech and the target speech vector; the drive of the vector length of the late parameter is used to evaluate and The driving sound source vector that becomes the smallest from the aforementioned driving target to the second target language of S becomes the smallest from the aforementioned distortion The source signal of the length, the other source codebook, uses M to generate the source vector; and the second frame sound uses M to generate the second round frame length code decoding device, which generates the vector length from the voice input on the encoding side The codebook is used to generate a vector source search mechanism, which uses K to evaluate the driving source of speech synthesis to the target of S that is the smallest to the source signal from the distortion to the smallest drive. On the other hand, the book uses M. Generate the corresponding delay; and the frame sound source to generate the sound source signal of the length of the frame. On the encoding side and the source sound source vector from the sound source encoding sound source vector to the sound direction 1 phase; and the first driving sound source should delay the reference source on the one hand The sound source signal of the generator is available, and the driver of the target length of the degree is adapted to the book, or the structure of the two-circle vector decoding number of the pair, the number of the target voice dynamics and the relative amount of the driven voice vector; and The picture frame moving sound source to Dong Zhong has a vector length with a parameter on the decoding side, which is driven by the sound (please read the precautions on the back before filling this page). Applicable to China National Standard (CNS) Α4 specification (210 X 297 mm) 16 1 β-Printed by the Ministry of Economic Affairs, Central Bureau of Standards and Engineering, Zhengong Consumer Cooperative A7 B7 V. Description of Invention (u) [Implementation of Invention] 霣 施Form 1. B1 is a block diagram of the speech «code device and speech decoding device of embodiment 1 of the present invention. In 圔 1, 1 is the encoder, 2 is the demultiplexer, 3 is the multiplexer, 4 is the sub-knock structure, 5 is the voice input, 6 is the pass line, and 7 is the voice_out ° coder 1 series It is composed of 8, 9, 15, and 25 ~ 36 under W. 8 is the nickel pre-parameter analysis structure, 9 is the linear pre-parameter parameter mechanism. 15 is the sound source signal generation mechanism, and 25 is the extracted voice input. The pitch analysis of the pitch period of the eaves period, 26 is the extended parameter search model of the retrieval model that determines the Ting Chi parameters when I retrieve the sound source vector, and 27 is the Pojia sampled speech input The speech key input spectrum plus sampling mechanism, 28 is the target speech generation mechanism that uses K to generate the target speech vector corresponding to the length of the IS parameter. The 29 is to increase the sampling of the sound irrigation signal generated in the past. Sampling organization · 30 is the self-supporting audio source codebook for outputting the self-entangled sound source vector of the vector length of the delay parameter from the yinhai information generated in the past · 31 is the self-supporting sound source retrieval structure · system evaluation and The aforementioned speech vectors of the speech synthesis vectors obtained from the sound source from Biaoying to Dong Zhong are relative to Search and search. The source vector with the least distortion is the self-supporting sound source vector. 32 is the ring frame sound source generating mechanism, which generates a frame length sound source signal from the self-paralysis sound source corresponding to the vector length of the extension parameter to ft. In addition, 33 is the structure of the second vocalization of the second ear, which is used to correspond to the most vocal glucoside of the vector length of the number of sound sources in the search of the ft to the ft; 34 is the drive of the sound source. Corresponding to the longest driving sound source vector in the direction of number # 35, 35 is the driving sound source eaves. The paper size is applicable to the Chinese National Falcon (CNS) A4 specification (210X297 mm). 17 —I- — &gt; i Bmli n L . ^ — A ^ i · ϋ ^ _— 1 · ϋ mmmm I (please read the note $ item on the back first and then fill in the ranunculus) Ordering line · 317631 Central Bureau of Standards of the Ministry of Economic Affairs * Printed by the Consumer Cooperative Cooperative A7 B7 5 3. Description of the invention (15) "Bangu 'uses M to evaluate the distortion relative to the second-order speech vector of the speech repertoire vector obtained from the moving sound source vector, and the search distortion becomes the smallest moving sound source vector, 36 Generate a structure for the second frame sound source, from the length of the corresponding delay parameter to 1 Those driving sound source signals use M to generate a frame-length moving sound source signal. In addition, the debouncer 2 is composed of the following 16, 21, 22, and 37 to 43. 16 is the linear prediction parameter deconstruction structure, 21 is the audio signal generation conversion structure, 22 is the synthesis filter * 37 is the Po source sampling the source signal generated in the past to increase the source signal sampling mechanism, 38 is the output delay parameter The number of vector lengths of the self-entangled sound source vector is self-adhesive, and the sound source tweezers are thin. 39 is the self-contained sound source decoding structure. 40 is a ring frame sound irrigation generating mechanism, which is from the self-selected Yingya to the suspension of the vector length corresponding to the delay parameter, and uses M to generate the sound source signal of the ring frame length. 41 is the thin brick that drives the sound source and outputs the corresponding delay. The vector length of the audio source driving the audio source vector is 42. It is the structure of the audio source decoding decoding. It is the signal source driving the source of the vector length corresponding to the delay # number * 43. It is the structure of the second garden frame sound generation. For the driving sound source signal of the vector length of the delay parameter, a K is used to generate a moving sound source signal of circular frame length. The action will be described below. First, in the encoder 1, for example, a digital speech signal sampled at 8 kHz is input as a speech input 5. The media pre-congestion parameter analysis structure 8 analyzes the aforementioned speech round 5 and extracts the media prediction parameters as the frequency envelope information of the speech. Next, the linear pre-holding I parameter configuration 9 quantizes the extracted linear pre-holding parameters above and compiles the corresponding _ to the multi-work paper size using the Chinese National Standard (CNS) A4 specification (2 丨 〇 &gt; &lt; 297mm &gt; -18-(Read the notes on the back first and then fill in this page) Packing & Threading * Α7 Β7 Printed by the Consumers ’Cooperative of the Central Bureau of Economic Development of the Ministry of Economy V. Invention Instructions (16) 3. At the same time, Dong Zihua ’s media prediction parameters are simultaneously output to the adaptive. Yinya search structure 31, the second-order speech generation mechanism 33, and the dynamic sound source search suspect structure 35. The pitch analyzing mechanism 25 5 analyzes the voice input 5M and extracts the pitch pass period P. Next, the delay # number search range determines the structure 26 from the aforementioned pitch pass period P, for example, according to formula (1) to determine the search parameter group 1 of the delay parameter 1 when retrieving from the episodic sound source vector 1 »t ηέΐέl» a X * The output is input to the voice round sampling sample structure 27, the sound source signal is increased by the sample structure 29, and the self-responding sound source retrieval mechanism 31. Here, ΔP is set to P / 10, for example. 1 m I η = Ρ-Δ Ρ 1 · a χ = Ρ + ΔΡ (1) The voice input increases the sampling mechanism 27 series M according to the delay parameter retrieval Fan Xing determines the structure of the retrieval parameters of the round 26 The sampling rate, for example, increases the sampled speech input 5 · to the target frame of the target speech generation structure 28 in the sigmoid frame interval which is the source of the audio signal. Here, the sampling rate is determined by, for example, M. In the case of 丨, increase sampling by 4 times. In 45 helmet lmtn &lt; 65 case * Double sampling is performed. In the case of 65 forget 1 «I η, Po plus sampling is not conducted. The target speech generating structure 28 is a voice input with an increased sampling round frame length input from the aforementioned voice input to increase the sampling sock structure 27, and corresponds to the delay parameter 1 input by the self-supporting sound source retrieval mechanism 31, for example, Μ Each period 1 is divided, and the length of the division is added to the speech input of the length of each segment by adding the average «parameters to the ft length. Η generates a vector length corresponding to the extension of the parameter 1 (please read the back side first (Notes and then fill out this page) h-installed. The size of the paper is applicable to the Chinese National Standard (CNS) A4 specification (210x297 mm) 19 317631 Printed by the Ministry of Economic Affairs Central Bureau of Accreditation Consumer Cooperatives A7 __B7_____ V. Description of invention The voice of (Π) 's head is in the direction of 5, and it comes out to the self-entangled sound source retrieval mechanism 31 and the second head's voice generation structure 33. Here, in addition to taking integer values, the delay parameter 1 can also take non-integer rational numbers. According to the existence paradigm of 1, for example, when It n t is embedded as an integer, the following values can be taken. at 1 &lt; 45 case, It n t, li n t + 1/4 * 1 · „1/2 * 1 t„ t +3/4 at 45 forget 1 &lt; 65 case, hnt, li nt +1/2 In 65Sil case, llnt circle 2 shows the most suitable example of the ft length of the target voice corresponding to the extension parameter 1 generated by the voice key input of the round frame length . Here, in the case of a generalized frame length, the above-mentioned addition averaging is not performed, and She takes the round of the length of the frame as the target speech vector. The source signal increased sampling frame 29 »is only in the interval required for the self-entangled sound source search of the search field according to the delay parameter input by the delay parameter search structure determined by the aforementioned delay parameter search fan garden, and M corresponds to the aforementioned delay The sampling rate of the retrieval parameter of the parameter is added to the sound source signal generated in the past input from the source structure 15 generated by the sound source signal, and the output signal M is output to the self-adapted sound source editing book 30. Here, increasing the sampling rate is determined, for example, according to the following. at 1 In the interval of &lt; 45, 4 times up sampling is performed. At 45 guests 1 In the range of &lt; 65, double sampling is performed. Between 65 ^ 1 S, no sampling is performed. The self-parametric audio source codebook 30 is derived from the up-sampled audio source signal rounded by the audio source signal crucible plus sampling structure 29, and will correspond to the direction of the delay parameter 1 input by the self-supporting audio source retrieval broadcast structure 31. The length of the ft should be from the sound source to the stupid (please read the note $ item on the back and fill in this page) η Order &gt; »T- 錬. This paper size is applicable to the Chinese National Standard (CNS) A4 specification (21〇X297 mm )-2〇 The Ministry of Economic Affairs, Central Bureau of Accreditation and Employee ’s Consumer Cooperative Printed «. A7 B7_ V. Description of Invention (18) ~ _To the adaptive audio source search structure 31. Here, the aforementioned self-sounding audio system cuts out a sampled audio signal from the delay parameter 1 and will cut out the audio signal of the frame length from a sampled past in the case of 1 degree of fear. The H sound source search mechanism 31 is equipped with a synthetic waver, and uses the quantized linear prediction parameter M input by the digital tweezers 9 to obtain the pulse response of the synthetic waver. Secondly, for the delay parameter of the llB t „彡 1 彡 1» ax, use the aforementioned pulse response response to calculate and repeat from the self-supporting audio source input by the aforementioned self-supporting audio source codebook 30 to W Generate the speech synthesis direction S. Next, find the target speech vector input from the narration target speech generation 28 and the auditory weighted distortion of the aforementioned speech synthesis direction ft. Second, compare and evaluate the aforementioned distortion, M find that the aforementioned distortion becomes minimum The extension parameter L of the corresponding and the corresponding self-supporting source point * and output the above-mentioned delay parameter L and the self-supporting source gain 々 yin code to the multiplexer 3 and the driving source tweezers code sheet 34, and generate The self-corresponding sound source parameter corresponding to the aforementioned delay parameter L is multiplied by the self-corresponding sound source signal of the aforementioned self-corresponding sound source gain / 3, and is output to the frame sound source generating structure 32 and the second target speech generating structure 33. Here, the aforementioned adaptive sound source signal is L-sampling in the case of L <frame length, and the circle-frame length in the case of LS_frame length. The circle-sound vocabulary generation structure 3 2 is derived from the foregoing Retrieval of self-supporting audio sources 31 In the input self-adapted sound source signal, for example, M is repeated and embedded in each pass period. Κ generates a self-adapted sound source signal with a frame length and outputs it to the sound source signal generation mechanism 15. This paper size is applicable China National Standard (CNS) Α4 specification (210X297 mm) -2 \-(please read the $ item on the back and then fill in this page)-Install · Threading and Printing A7 B7 5. Description of the invention (19) The second target speech generating mechanism 33 uses the quantized linear prediction parameters input by the media prediction parameter Zen code mechanism 9 to retrieve the self-inputs from the aforementioned self-sounding sound source retrieval structure 31 Adapt the sound source signal to perform linear prediction synthesis M to generate a speech synthesis vector. Next, find the difference between the target speech vector input by the Meg speech generation mechanism 28 and the aforementioned speech synthesis vector, and output it as the second target speech to 1. To the driving sound source retrieval mechanism 35. On the driving sound source codebook 34, for example, N driving sound sources generated from the noise are stored to 1, corresponding to the vector length of the aforementioned delay parameter L The output and output correspond to the driving sound source code i of the driving sound source code i input by the driving sound source retrieval structure 35. Here, in the case where L is the length of the frame, she rounds out the driving sound source of the round frame length to Dong. Retrieve the 35th pair of N sound drive sound sources to M, and use the ft sub-sources input by the aforementioned drive sound source codebook 34 to use the ft input by the aforementioned media prediction parameters to the code 9 Linear prediction parameters are used to perform linear prediction synthesis M to generate a speech synthesis vector. Next, the auditory addition distortion of the second target speech inputted by the second target speech generation mechanism 33 to Dong and the aforementioned speech synthesis vector is obtained. Secondly * comparison Evaluate the aforementioned distortion, find the driving sound source code I and the corresponding driving source gain 7 corresponding to the minimum distortion, and output the codes of the driving sound source code I and the driving sound source 7 to the multiplexer 3 * At the same time, a driving sound source signal corresponding to the aforementioned driving sound source code I is multiplied by ft before K_driving sound source gain 7 and is output to the second frame sound source generating structure 36. The second frame sound source generating mechanism 36 is based on the driving sound source signal input from the aforementioned driving sound source retrieval mechanism 35, for example, the Lit line is repeated every cycle, and the paper scale is adapted to the Chinese National Standard (CNS) A4 specification (210X297 mm ) -2 2-(please read the $ item on the back and then fill out this page) -install. A7 B7 printed by the Central Bureau of Economics and Trade, Beigong Consumer Cooperative of the Ministry of Economics. 5. Description of invention (20) Buried, The driving sound source signal of the length of the circle frame is generated and output to the sound source signal generating eaves 15. The sound source signal generating structure 15 is a summation of the self-tuning sound source signal input by the aforementioned plaque frame sound source generating mechanism 32 and the driving sound of the garden frame length input by the aforementioned second round frame sound source. The ya signal K generates a ya signal and outputs it to the sound source signal up-sampling mechanism 29. After the end of K-Yong, the multiplexer 3 sends out the codes corresponding to the aforementioned quantized linear pre-spray parameters, and the codes corresponding to the delay parameter L, the driving sound source Burmese I, and the sound source «益 / 3, 7 To transmission medium 6. Μ 上 is a special action of the voice code device of the first form. Secondly, the operation of the decoder 2 will be described. "First, the separation cycle 4 of the output of the multiplexer 3 is accepted, and the Y code of the linear prediction parameter is separately output to the linear pre-parameter parameter decoupling mechanism 16; the delay parameter The tweezers code of L is output to the self-restrained sound source decomposing mechanism 39 and the hurricane sound source metal code thin 41; the sound source gain Ling Zhige is output to the self-supporting sound source decoding structure 39; and the sound source sound source I The sound source of the sound source 7 is rounded out to drive the sound source decoding mechanism 42. The self-supporting audio source decoding structure 39. The system first outputs the aforementioned embedding parameters to the audio source signal plus extraction mechanism 37 and the self-supporting audio source. The audio signal is added with a rubbing mechanism 37. Between the plaques required for generating the self-responsive sound source vector according to the value of the aforementioned delay parameter L input from the self-defining sound source decompression structure 39, K follows the aforementioned delay parameter Sampling rate of the value of L * Pu plus samples the vocal sink signals generated in the past from the Yone Ma signal generation Yong structure 21 for output to the audio source audiobook 38. Here • Increase the sampling (please read the note $ item on the back and fill in this page). Binding. Threading-This paper size applies to China National Standard Vehicle (CNS) A4 specification (210 X 297 mm) 23 317631 Α7 Β7 Printed by the Ministry of Economy Central Bureau of Accreditation Consumer Cooperatives V. Description of invention (21) The rate system and the sound source signal in encoder plus sampling structure 29 are also decided. The self-responding sound source codebook 38 is from the up-sampled vocal ensemble signal rounded by the aforementioned sound source signal up-sampling structure 37, and will correspond to the delay parameter L rounded by the self-entangled sound source decoding structure 39. The vector-length self-edge response sound source is output to the self-edge response sound solution structure 39. Here, the person who responds to the audio source cuts the delay parameter L to the sound source signal passing through the L parameter, and in the case of the frame length, cuts the sound source signal with the length of the frame from the L sample. The self-recording audio source decoding mechanism 39 is to extract the self-recording audio source Poyi / 3 from the aforementioned self-recording audio source Yiyi Lengyi, to generate the self-recording audiobook input codebook 38 The Yuning sound source outputs the self-supporting sound source signal which is the most multiplied by the aforementioned K from the Yunyi sound source gain Lu, and outputs it to the H frame sound generation generating eaves structure 40. The frame sound source generating mechanism 40 is a self-contained sound source signal input from the self-contained sound source decoding mechanism 39, for example, κ is repeated for each buried period, and K generates a self-contained sound source signal with a frame length, and _ Out to the sound source signal generating mechanism 21. The audio source codebook 41 stores the same N-driver sound source vector as the drive source codebook 34 on the coding side, and cuts out the length of the direction fi corresponding to the aforementioned delay parameter L to be decoded by the SI dynamic sound source. The driving sound source vector of the input driving sound source 颃 碣 I is output to the driving sound source decoding and broadcasting structure 42. The driving sound source decoupling mechanism 42 decodes the driving sound source gain from the aforementioned code of the moving sound source gain 7, and M generates a cut-off body sound source vector rounded by the driving sound source level code thin 41 multiplied by the aforementioned sound source Po The sound source signal of Yi 7 is driven out to the second ring frame sound source generation broadcast structure 43. The second _ frame paper size is applicable to China National Standard (CNS) Α4 specification (210Χ 297 mm) ----- J ---- I installed-- (please read the precautions on the back before filling this page) A7 B7___ Printed by the Ministry of Economics of the Ministry of Economics, Central Bureau of Rectification and Consumer Cooperatives 5. Description of the invention (22) # is to generate the structure 43 from the sound source signal input by the aforementioned sound source decoding structure 42, for example A buried period P is repeatedly encountered to generate a driving sound irrigation signal of a round frame length, and then rounds out to the sound source signal generating structure 21. The sound source signal generating mechanism 21 adds the adaptive sound source signal of the H frame length input by the aforementioned frame sound source generating structure 40 and the length of the frame frame input by the aforementioned second frame sound source generating structure 43 The dynamic sound source signal M generates a sound source signal * and outputs it to the sound source signal plus sampling structure 37 and synthesis filter 22. The synthesis filter 22 is a sound source signal input by the aforementioned sound source signal generation mechanism 21, using The linear prediction parameters keyed in by the parameter unblocking mechanism 16 perform media prediction synthesis, and output voice output 7. The above is the extreme action of the voice decoupling device of the first embodiment. According to the first embodiment, when determining the optimal embedding parameter, when the prolongation parameter 1 is shorter than the frame degree, the average speech input is added periodically to generate the speech vector of length 1 Relative to this estimate and linear prediction of the synthesis vector, the distortion of the speech synthesis vector generated from the Yingyinya vector from the first degree of the synthesis vector county • and * even when deciding on the best source of the sound of movement " Distortion evaluation uses the media prediction synthesis to generate the speech synthesis vector generated from the driving sound of Dong length 1 to fi, to avoid the deterioration of the quality of speech synthesis, and to generate speech synthesis with low cost and good quality. Ensuring Form 2. In the above-mentioned Benshi Form 1, the _frame sound source generates "frames 32, 40 and the second country frame sound source generating mechanism 36, 43, although it is the same for each period of L. It will correspond to the delay. The parameter of the length of the parameter L to the fi length of the self-cockroach signal or the sound of the sound book is suitable for the Chinese national standard (CNS & A4 specifications (2 丨 0X297mm) -25-(please read note f on the back (Fill in this page again) η -installation. 317631 A7 B7. V. Description of the invention (23) When the source signal meets the period, K generates an adaptive sound source signal or driving sound source signal of the length of the garden frame, but it can also correspond to the aforementioned delay. The adaptive sound source signal or driving sound source signal of the length of the parameter L, for example, M performs waveform interpolation for each pass L, interpolates between the circles, and M generates a frame length adaptive sound source signal or driving sound source signal. According to the second form of the trade application, the change of the sound source signal between the circles becomes smooth, the resynthesis of speech synthesis is good, and the quality is improved. Embodiment 3. In the above-mentioned Embodiments 1 and 2, Although it is an adaptive response from the corresponding delay parameter L to the length of Dong In the source signal or the driving sound source signal, use the ring frame sound source to generate the frame and the second frame sound source to generate the frame K to generate the frame length of the adaptive source signal or the frame length of the driving source signal, and add the same Generate a sound source signal of frame length, but you can also add an adaptive sound source signal of the vector length corresponding to the delay parameter L or drive the sound source signal to generate a sound source signal of the vector length corresponding to the delay parameter L. For example, each cycle L will repeatedly Encounter the period, and generate a sound source signal of the frame length. G Application form 4. Printed by the Beigong Consumer Cooperative of the Central Bumper Bureau of the Ministry of Economic Affairs (please read the precautions on the back and then fill out this page) In the first embodiment above Although both the encoder and the decoder adopt new configurations, the encoder can be used as the encoder of Embodiment 1 and the decoder as the conventional decoder described in FIG. 12. Embodiment 5. In the above In the first embodiment, the target speech generating mechanism 28 generates the target speech vector corresponding to the vector length of the delay parameter 1 from the speech input of the national frame length, but it can also be As shown in 3, the Chinese national standard (CNS) A4 specification (210 X 297 mm) is applied from the corresponding delay reference paper scale -2 6 _ 317631 A7 ._____'_ B7_ V. Description of invention (24) The number of vector length of number 1 The speech vector. The speech vector is generated from the speech input of an integer multiple length. (Please read the note $ item on the back and then fill in this page.) According to the fifth embodiment, in the averaging process when generating the target speech vector, there is no If vectors with different vector lengths are necessary, it can be easily processed. Furthermore, by using it for evaluation when the speech input exceeding the frame length is subjected to speech dumping, the reproducibility of speech synthesis can be improved, and The quality is improved. Embodiment 6. In the above-mentioned Embodiment 1, the target speech generating means 28 is simply averaged when generating the target speech vector corresponding to the length of the delay parameter 1 from the voice input to the fi length, but it is also As shown in IB 4, the power input of the voice input according to the vector length corresponding to each delay parameter 1 can be used. For example, the higher the power, the larger the added broadcast, and the average is added. According to the sixth embodiment, in the averaging process when generating the target speech vector, by performing an on-air speech coding using a portion with a large input power of speech, speech synthesis with a large influence on subjective quality can be achieved The Central Ministry of Power Economics has printed high quality products for the Beigong Consumer Cooperatives to improve the quality and make it into a transgender regulation. Then the delay of the real large and medium number parameters applied by the fractal department should be dealt with into the ecological form. From the 28-degree measuring machine in the system, the long-term measuring machine to the Chengshengsheng I, the pronunciation of Baibai is in the use of Poligo's rate indicator, It enters 0, the length of the speech is 5 degrees, and the amount is pure U. The singular reference should be delayed for each of the delays M should be according to the picture according to the corresponding pronunciation, but the long number of the input speech should be included in the delay. , As close to the example, enter 27 paper standards using the Chinese National Standard (CNS) Α4 specification (2! 0 &gt; &lt; 297 mm) 317631 A7 B7 Printed by the Ministry of Economic Affairs, Central Bureau of Economic Development, Jeonggong Consumer Cooperative V. Description of the invention (25) 1 1 If the target speech vector is generated according to the seventh embodiment, the average 1 1 average In the processing, the speech input is coded by reducing the weight of the low-correlation part of the 1 1 case with the periodicity of period 1. Even for the pitch cycle, please change the voice input by 1 1 period first to generate the corresponding 1 section. The visual distortion of the 1 1 mark speech vector with less distortion from the period improves the reproducibility of speech synthesis and improves the quality. Note 1 to the implementation form 8.1 Item μ In the above-mentioned embodiment 1, the target speech generating mechanism 2δ is filled in with 1 as the target% of the length of the direction corresponding to the delay parameter 1 is generated from the speech input. 1 The speech vectors are simply averaged, but they can also be shown in Figure 6 * Use 1 1 to weight the power of the speech input corresponding to the length of each delay parameter 1 to 1 1 I »For example, for speech input near the border of the frame The weighting is increased to perform phase 1 1 order addition average 0 1 I. According to the eighth embodiment, the average 1 1 homogenization process at the time of generating the target speech vector is performed by adding in the speech input near the boundary of the ΓΕΠ brain frame A large weighting of 1 to generate a huge τη preserving speech direction 3 and encoding it can make the speech synthesis near the border of the frame γ line | have good reproducibility, and can make the change of speech synthesis between frames become smooth 0 The effect is between the frames Especially significant when generating interpolation Embodiment 2 in the case of a sound source signal. 1 1 spring application form 9. 1 1 In the above-mentioned first embodiment, in terms of the target speech generating mechanism 28, although 1 I is like generating a target corresponding to the length of the delay parameter 1 from the speech input, although the length of the target 1 1 I is At the time of M, the average speech input is added every 1 cycle, but it can also be 1 1 I as shown in FIG. 7, for example, in order to make the speech length corresponding to the vector length of each delay parameter 1 1 1 sound input related (c Γ 0 SS -correlation) becomes the largest • and micro 1 1 This paper scale is applicable to the Chinese National Standard (CNS) A4 specification (210X297 mm) -28- A7 B7 Printed by the Ministry of Economic Affairs Central Bureau of Unemployment Consumer Cooperative «. V. Invention Explanation (26) Li cut out the position where the voice turns in * and add them to the average. According to the real pickling body 9, in the averaging process when generating the vocal speech direction 1, in order to maximize the mutual gap between the speech wheels corresponding to the length of the direction of the corresponding parameter 1, the fine-tuned cut is cut out Position, even if the round of the pitch of the period changes, it can generate a small distortion of the corresponding speech vector corresponding to the pitch of the period, so that the speech synthesis has good reproducibility * and improves the quality. Information form 10. Picture 8 is a block picture showing the full-level structure of the voice yoke coding device and voice decoding device of the cell form 10 of the present invention. In this circle, the same part as circle 1 is given the same symbol, and its explanation is omitted. In H8, the new structure compared to circle 1 is shown below. 44 is a voice input to increase the sampling voice input and a rubbing mechanism is added, 45 is a mute voice generating eaves structure that uses M to generate a sample voice vector corresponding to the vector length of the pitch encounter period, and 46 and 51 are thin body sound sources. , Is to output the driving source corresponding to the length of the pitch ft to the ft. 47 is the retrieving and broadcasting structure of the moving source. It is used to evaluate and synthesize the speech synthesis from the moving source to JI. The relative distortion of the speech vector, and the retrieval distortion becomes the smallest driving sound sea vector, 48 generates a structure for the second target voice. It is used to generate the corresponding second drive, the length of the pitch period in the sound source vector retrieval. For the target speech vector, 49 and 54 are the second driving sound source, which is the second driving sound source vector for outputting the vector length of the edit pitch, and 50 is the second broadcasting sound source retrieval mechanism. Used to evaluate the distortion relative to the speech synthesis vector obtained from the second sound source from Dong to the second-order speech vector of 1, and the search distortion becomes the smallest drive sound source to fi. 52 is the drive sound source solution mechanism. Paper ruler Applicable to Chinese National Standards (CNS & A4 specifications (210X297mm) -29-(please read the $ item on the back and then fill in this page) • Binding · Strapping. S17631 A7 __B7_ 5. Description of the invention (27) for The driver that decodes the vector length of the pitch corresponding to the pitch period. 53 is a group frame sound source generation mechanism. It is a driver who generates a ring length driving sound source signal from M from the drive pitch vector that corresponds to the vector length of the pitch period. 55 is the decoding structure of the second joyful sound irrigation system. It is the second driving sound source signal that decodes the length of the corresponding pitch through period corresponding to the length of the Dong. 56 is the generation of the tracking structure for the sound frame of the second frame. It is from the jT pitch. In the second driving sound source signal of the vector length of the period, it is used to generate the driving sound source signal of the frame length. Below, the operation will be described centering on the above-mentioned new structure. First, in the «coder 1, the pitch analysis supports The system 25 analyzes the voice input 5 and extracts the pitch buried period P to output to the multiplexer 3, the voice input increase sampling mechanism 44, the target voice generation and broadcast structure 45, the driving sound source codebook 46, the second driving sound source code Thin 49. Here, the pitch pass P value other than an integer, non-integer rational also taken according to the presence of P * Fan, e.g. Μ acquired value of the "Spanish fly in the case where the set value * of the pitch of the through At Ρ &lt; 45 case, Pi n t »Pl n t + l / 4, Pl n t ♦ 1/2. P I n t + 3/4 at 45 forget P &lt; In case of 65, it is P tn *, P tnt +1/2 Printed by the Beigong Consumer Cooperative of the Ministry of Economic Affairs of the People ’s Republic of China (please read the notes on the back before filling in this page) In the case of 65SP, it is Pint The voice input increasing sampling mechanism 44 is based on the extraction rate of the pitch pass period entered by the pitch analysis mechanism 25. For example, the sampled voice is added to the round frame interval which is the position of the source code of the code. The output is to be output to the Miao speech generation mechanism 45. Here, the Poka sample rate is determined by, for example, the following. In Ρ In the case of &lt; 45, increase sampling by 4 times. At 45SP &lt; 65 times, buries twice as much as rubbing. This paper uses the Chinese National Standard (CNS> A4 specification (210X297mm) -3〇 ** Printed by the Ministry of Economic Affairs of the Ministry of Economic Affairs, Shellfish Consumer Cooperative A7 ___B7_ 5. Description of the invention (28) at 65SP No additional sampling is performed. The vowel speech generating mechanism 45 rounds up the speech of the frame length of the Po plus sampling that was input by the aforementioned speech input increasing sampling mechanism 44, corresponding to the pitch period input by the pitch analysis mechanism 25 , And by, for example, adding and averaging each buried period p, K generates the head voice vector of the vector length P, and the key is output to the II moving sound source retrieval structure 47 and the second head voice generation mechanism 48. Here, In the case of P ^ · frame length, the aforementioned addition and averaging are not performed, and the speech input of the frame length is used as the head voice to ft. On the driving sound source yin milling 46, there are, for example, those generated by the late sound of sugar eaves The N sound sources are suspended, using the pitch corresponding to the input from the pitch analysis mechanism 25. The vector length of period P is cut out and the output corresponds to the sound source code input by the waist motion source checker structure 47 i's moving sound source vector. Here, in P In the case of the length of the circle frame, the boundary motion source vector of the length of the frame is output. The driving sound source retrieval mechanism 47 uses the driving sound source vector cut out from the input of the ancestral moving sound source code code 46 for the N ancestor moving sound source vectors The linear prediction synthesis is performed on the quantized intermediary prediction parameters of the entangled prediction parameter 9 compiled by the eradicative prediction parameters to generate the speech synthesis direction 1. Next, the rounds are obtained by the pre-sentence speech generation mechanism 45 The target speech vector and the speech synthesis vector's auditory add-on distortion. Secondly, the aforementioned distortion is compared and evaluated to obtain the dynamic sound source code I and the corresponding drive _ sound source heap 7 that the aforementioned distortion becomes the smallest. The aforementioned driving sound source code I and the driving sound source gain 7 are outputted to the multiplexer 3, and at the same time, the driving sound source corresponding to the aforementioned driving sound source I is generated by multiplying the driving sound source signal ft by the aforementioned driving sound source gain 7 and output To the second item, "Speech Generation 48", this paper size is applicable to China National Standard (CNS) A4 (210X297mm) -31-(please read note f on the back and fill in this page) h Line ' A7 B7__ Printed by the Consumer Labor Cooperative of the Central Bureau of Economic Development of the Ministry of Economy 5. Description of the invention (29) The second-generation voice generation unit 48 uses the driving sound source signal input by the aforementioned driving sound source retrieval structure 47 to use linear pre-I The parameterized quantized medium prediction parameters entered in the structure 9 are subjected to linear prediction synthesis to generate a speech synthesis vector. Next, the target speech vector and the aforementioned speech input by the aforementioned structure generating structure 45 are obtained The difference of the synthesized vectors • It is output as the second target speech vector to the second sound source retrieval structure 50. On the second driving sound source code book 49, for example, a random voice is generated. N tilt drive sound source. Use the direction ft length corresponding to the pitch period P input by the pitch analysis mechanism 25 and _ corresponding to the drive sound source code j input by the drive sound source retrieval mechanism 50. The second sound source vector. Here, in the case where P is at the length of the frame * the driving source of the frame length is rounded to 50. The second moving sound source retrieval mechanism 50 sets N moving sound source vectors, which will be thinner by the second driving sound source. The second sacral motion sound source vector entered in 49 is subjected to media prediction synthesis using the linear prediction parameters input by the aforementioned linear prediction parameter editing mechanism 9 to generate a speech synthesis vector. Next, the sleep-on-play distortion of the second-meat speech vector and the speech-synthesis vector entered by the second-mean speech generating unit 48 are obtained. Secondly, the aforementioned distortion is compared and evaluated to obtain the second driving sound source Burr J that has the smallest distortion and the corresponding driving sound sea gain 7 2, and the aforementioned second driving sound source Tweezer J and driving sound source gain 7 2 Burmese yards round up to the multiplexer 3 °

After the above code is over, the multiplexer 3 will make bricks corresponding to the aforementioned quantized linear prediction parameters, and correspond to the pitch period P and the radiating sound source. This paper scale is applicable to the Chinese National Standard (CNS ) A4 specification (210X297 mm) · 32- (please read the $ item on the back and then fill in this page) h-line booking-A7 B7 printed by the Beigong Consumer Cooperative of the Central Bureau of Economic Development of the Ministry of Economic Affairs (3〇), J and sound source gain 7, 72 are sent to the code Fu Fudai 6. The above is the special action of this phonetic code device for Shi Xingjian 10. Next, the operation of the decoder 2 will be explained. Stomach &amp; accepting the multiplexer 3, the separation mechanism 4 * separately outputs the linear pre-code ## 5 * ϋB code to the linear prediction parameter decoding structure 16; outputs the code of the pitch «period P to the drive The source rein codebook 51, the second driving sound source is thin [5 4; turn the driving sound source code I, the source gain 7 tilt code wheel out to the driving sound source decoding structure 52; and the second driving sound source "« J. The sound source «Yi 7 2 Burmese code is outputted to the second moving sound source decoupling mechanism 55. ϋ The codebook 51 for moving sound source stores N number of metaphonic sound source vectors that are the same as the driving sound source for the code 46, and Μ corresponds to the length of the preceding pitch cycle ρ. The drive sound source vector of the drive sound source code I input by the decoding structure 52 is output to the drive sound source decoding structure 52. The driving sound source decoding structure 52 is to extract the driving sound source crucible 7 from the metal sound of the aforementioned driving sound source benefit 7 to generate a driving sound source which is cut out by the aforementioned fraudulent sound source 碖 碣 thin 51. The driving sound source signal multiplied by the aforementioned sound source gain 7 is output to the ring sound source generating structure 53. The plaque frame sound source generation structure 5 3 is derived from the fraudulent sound source information input by the aforementioned moving sound world decoding broadcast structure 52, for example, it is repeated in stages with each pass period, Μ generates _ frame length driving sound source The signal is output to the audio signal generation module 21. The second driving sound source codebook 5 4 series has the same K body sound source vectors as the second driving sound source codebook 49 on the coding side, and uses The vector length corresponding to the aforementioned pitch pass period P is cut out to correspond to the second moving sound source vector input by the second driving sound source decoding structure 55 inputted by the second driving sound source 颛 碣 J ' (CNS) Α4 specification (210x297mm) -33-(please read the note on the back first and then write this page) -installation. Stranding 317631 A7 B7 5. Description of invention (31) Yizengyuan Audio Drive II Before the first statement. On the system, 5 5 5 5 construct the source code solution to decode the source source audio iIS, the second to second decode the second waist audio source printed by the 2nd to the Ministry of Economic Affairs Central Sample Bureau Employee Consumer Cooperative 7 2 Gain 72, K generates a second drive sound source signal cut out from the second drive sound source codebook 54 to the Dong multiplied drive sound source gain 72, and is output to the second The second frame sound source generating mechanism 56 is provided. The second ring frame sound source generating mechanism 56 is derived from the second driving sound source signal rounded by the aforementioned second meta-motion sound source decoding structure 55, for example, Μ is repeated for each buried period to generate a second length Drive the sound source signal, and output to the sound source signal generating mechanism 21. The sound source signal generating structure 21 adds the self-responsive sound source signal of the length of the face frame input by the aforementioned round frame sound source generating mechanism 53 and the second round frame input by the aforementioned second round frame sound source. The sound source signal of the length is driven to generate a sound source signal, and output to the synthesis filter 22. Synthesis »The waver 22 is to linearly predict and synthesize the audio signal input by the aforementioned sound source signal generating mechanism 21 using the linear prediction parameters input by the linear pre-owning parameter decoding mechanism 16, and to output the voice 7 . The above is the extreme action of the voice decoupling device of the tenth embodiment. If according to the present spring application form 10, the average speech input is added by the periodical addition when the pitch of the speech input is shorter than the length of the circular frame to generate the speech of the length of Dong to the Dong, which is relative to This evaluation and media prediction predicts the distortion of the speech synthesis vector generated by the driving sound source of the synthesis vector length P to fl * It can avoid the deterioration of the quality of speech synthesis and can generate speech synthesis with less calculation fi and good quality. [Effects of the invention] (Read the note $ item on the back first and then fill in this page) • Binding ·-Threading-This paper scale is applicable to the Chinese National Standard (CNS) A4 specification (210X297 mm) -34 317631 A7 _ B7_ 5 Description of the invention (32) As detailed above, according to the inventions described in items 1 to 4, 6, 8, 9, 9 and 11 to 14 of the patent application scope, since the voice encoding device is provided with: The target speech generating mechanism generates a target speech vector corresponding to the vector length of the delay parameter from the voice input; the adaptive sound source codebook generates an adaptive sound source corresponding to the vector length of the aforementioned delay parameter from the sound source signal generated in the past Vectors; an adaptive sound source retrieval mechanism that evaluates the distortion relative to the target speech vector of the speech synthesis vector obtained from the adaptive sound source vector * and retrieves the adaptive sound source vector with the least distortion; and the frame sound source generation The mechanism is to generate a sound source signal with a frame length from the aforementioned adaptive sound source with the smallest distortion, so that the quality of speech synthesis can be avoided, and the Speech synthesis into operation less and good quality. In addition, according to the invention described in item 5 of the patent application scope, although the length of the target speech to Dong Zhi is taken as a rational number, when K generates a target speech vector from the speech input, there is no reason for the input of the speech The sampling period can accurately generate the target speech to Dong, and can avoid the deterioration of the quality of speech synthesis, and can generate speech synthesis with less calculation and good quality. Printed by the Beigong Consumer Cooperative of the Central Bureau of Economic Affairs of the Ministry of Economic Affairs (please read the precautions on the back before filling in this page). Also, if the invention described in item 7 of the scope of patent application is applied, the voice generator Each vector length divides the speech input corresponding to an integer multiple of the vector length of the delay parameter, and adds the average of the speech input of each direction S length to generate the target speech direction S, so the average when generating the target speech vector In the process of processing, it is not necessary to deal with vectors with different vector lengths, it can be simply processed * and can avoid the deterioration of the quality of speech synthesis, and can generate speech synthesis with less calculation and good quality. This paper scale is applicable to the Chinese National Standard (CNS) A4 specification (210X297mm &gt; -35-317631 A7 B7 Printed by the Central Bureau of Economic Affairs of the Ministry of Economic Affairs Beigong Consumer Cooperative. V. Invention description (33) 1 If another giant standard will be generated The integer length of the vector length of the vector is Z. The length of the voice Ί 1 input is set to be longer than the frame length. It is used for evaluation when speech coding is performed on the voice 1 1 input that exceeds the frame length &gt; and by the frame Words 1 1 tone synthesis also adds to the impact of the frame M and decides the encoding can make reading 1 back 1 00 tone synthesis good reproducibility and improve the quality of 0 from the note 1 and if it is every time- * The feature quantity input to the speech of 1 length contains at least the meaning 1 ^-The power information of the button input is weighted by 4 1 in the part of the power of the speech input Tone coding can make language writing that has a great influence on subjective quality. This page contains 1 the reproducibility of the part with high power of sound synthesis and improves the quality. 1 1 If the characteristics of speech input per vector length The trace contains at least 1 I of the power information of the voice input. By reducing the weight of the relevant low part and performing voice coding when the voice input has a cycle of 1 of the cycle 1 1 schedule, the voice input of the pitch cycle changes It can also generate S 1 standard speech vectors corresponding to 1 pitch period with low distortion, which makes the speech synthesis reproducible and improves the quality of 1 1 0 l · line and if the standard speech generation generates im according to the length of each vector The time relationship between the input sound input and the average speech input 1 of each of the aforementioned vector lengths determines the weighting when generating the giant speech vector Then, by increasing the weight in the speech input near the frame border 1 1 to generate a hundred-standard speech vector and encoding 1 |, the reproducibility of speech synthesis near the frame border is good, 1 I can also make The change of speech synthesis between the frames is smooth 0 1 1 I and »If the giant speech generator generates the average language 1 1 sound input with each vector's degree of fear &gt; fine-tune the aforementioned speech of each vector's degree of fear The input time Μ 1 1 series »then by the vector length of the speech m between the input m becomes the largest and the micro 1 1 The paper size is applicable to the Chinese National Standard (CNS) A4 specification (210X297 mm) -36-Central Ministry of Economic Affairs A7 B7 printed by the quasi-bureau consumer cooperative. V. Description of the invention (34) Adjust the cut position. Even for the speech input with a variable pitch cycle, a target speech vector with a small distortion corresponding to a pitch cycle can be generated. Make the reproducibility of speech synthesis good, and improve the quality. In addition, according to the invention described in item 10 of the patent application range, since the frame sound source generating mechanism interpolates the sound source vectors between Dong Wei and the frame to generate the sound source signal, the sound source signal between the frames can be used The changes become smooth, making the speech synthesis reproducible and improving the quality. [Brief description of drawings] Fig. 1 is a block diagram showing the overall configuration of a speech coding apparatus and a speech decoding apparatus according to Embodiment 1 of the present invention. Fig. 2 is an explanatory diagram showing an example of the operation of the target speech generating mechanism in Embodiment 1 of the present invention. Fig. 3 is an explanatory diagram showing an example of the operation of the target speech generating mechanism in Embodiment 5 of the present invention. Fig. 4 is an explanatory diagram showing an example of the operation of the target speech generating mechanism in Embodiment 6 of the present invention. Fig. 5 is an explanatory diagram showing an example of the operation of the target speech generating mechanism of the funding form 7 of the present invention. FIG. 6 shows an explanation of the actions of the target speech generating mechanism implementing the state 8 of the present invention. Fig. 7 is an explanatory diagram showing an example of the operation of the target speech generating mechanism in Embodiment 9 of the present invention. Circle 8 shows a block diagram of the overall configuration of a speech coding apparatus and a speech decoding apparatus according to Embodiment 10 of the present invention. This paper scale is applicable to the Chinese National Standard (CNS) A4 specification (210X297mm) -37-(please read the precautions on the back before filling in this page) f-Binding · Stranding Winter 8a 8.: u A7 B7 Central Ministry of Economic Affairs Sample printing of quasi-bureaus and labor cooperation cooperation Du Du. V. Description of the invention (35) 1 1 I FIG. 9 shows a conventional speech encoding and decoding device-the overall configuration block of the example 1 1 I FIG. 〇1 I FIG. 10 (a) , (B) Temporary display of adaptive sound in the conventional speech encoding and decoding device, please first read the description of an example of I source vector. Figure 0 Read back | ίι I Figure 1 1 (a), (b) shows improved knowledge Note 1 of S in the speech coding and decoding device 1 1 Adaptation to the sound source vector-example explanation circle 0 Illness 1 I plaque 12 shows another difference of the conventional speech coding and decoding device and then fill in 1% of this device * example Block diagram of the overall composition 0 X 1 1 Figure 13 (a) '(b) shows the periodical 1 1 drive in the conventional speech m-code decoding device Audio source 1-quantity description-example illustration 0 1 1 C component number description] 1 order 1 encoder, 1 I 2 decoder, 1 1 I 3 multiplexer, 1 1 4 separation mechanism &gt;, 5 voice input | 6 transmission lines, 1 I 7 speech output 1 1 I 8 linear prediction parameter analysis mechanism, 1 1 9 linear prediction parameter coding mechanism &gt; 1 1 10, 17 self-adapted audio source coding thin t 1 I 11 adaptive audio source retrieval mechanism t 1 I 12 Error signal generation mechanism 1 1 I 13, 19 Drive source code thin t 1 1 This paper scale is applicable to China National Standard (CNS) Α4 specification (210X297 mm) ^ .Τ · ΓΓ 38 A7 B7 Central Ministry of Economics Printed by the Bureau and the Industry and Consumer Cooperatives 5. Description of the invention (36) 1 14 Driving audio source detection% Structure 1 1 I 15 '21 Audio source signal generation eaves structure 9 1 1 16 Linear measurement parameter decoding mechanism 9 Please first 1 1 18 Bai Yunying's sound source resolution mechanism »Read 1 Back 1 20 Note on the drive sound source resolution mechanism 1 22 Synthesizer m waver» I Item 11 rl 23 '24 Drive source m codebook * Re-install 1 25 pitch analysis eaves This page 26 The delay parameter search paradigm determines the eaves structure, 1 1 27 Voice input increase sampling mechanism 9 1 I 28 Hundreds of speech generation level structure I 1 order 2 9 '37 sound source signal increase sampling structure, 1 30, 38 from The sound source is very thin 1 1 31 I The sound source search structure * 1 32 '40 The frame sound source generation structure 1 line 33 The second giant purlin speech generation structure 1 I 34, 41 Drive source code book 1 35 Driven sound source retrieval mechanism 1 1 36 '43 second frame sound source generation structure, 1 | 39 im response sound source decongestion 1 I 42 drive sound source decoding structure, 1 1 I 44 speech input increase sampling structure • 1 1 45 Ju You Speech Generator »1 1 46 '51 driver sound source m code thin 9 1 1 This paper scale is applicable to China National Standard (CNS) A4 specification (210X297 mm) -39-317631 A7 B7 V. Invention description (37 ) Printed by the Central Bureau of Economics of the Ministry of Economic Affairs of the Consumer Cooperative 47 Driven Audio Source Retrieval Agency 9 48 Second Megaphone Speech Generation Organization t 49, 54 Second Driven Audio Source Code Book 50 Second Drived Audio Source. Retrieval Agency 9 52 Driven Audio Source Decoding Organization 1 53 Frame sound source generation mechanism 9 55 Second drive sound source decoding mechanism f 5 6 Second frame sound source generation mechanism 〇 (Please read the precautions on the back before filling this page)

A • Binding. Threading This paper standard is applicable to China National Standard (CNS) A4 (210X297mm) -40-

Claims (1)

Ministry of Economic Affairs * Rate ^ Zhengong Consumer Cooperative Printed A8 B8 C8 --- D8 VI. Patent application Fan ι · -Voice audio device, which is used to divide voice into audio frequency to envelop stomach messages Harmonize the sound source signal information, and use the _frame unit to code the sound source signal information of the stomach sound 1816 device. The special excitement is that it has: the structure of the head voice generation, &amp; &amp; stomach sound _ into the corresponding delay parameter The vector length of the head voice 5 is a self-entangled sound source, which is a sound source signal generated from the past into a self-considered sound source vector corresponding to the length of the aforementioned embedding parameter; the self-contained sound source retrieval agency * is evaluated The distortion retrieval relative to the speech synthesis vector obtained from the aforementioned lame response sound source vector + the aforementioned target speech vector to Dong is the one with the least distortion from the corresponding sound source to the ft; and the Li frame sound source generates the structure from the aforementioned distortion For the smallest self-grinding sound source vector, the sound source signal of the circle frame length is generated. '2. If the application of the patent Fanyuan No. 1 of the speech code device, it is further equipped with: a second target stack voice generation mechanism, which is generated from the target poke speech to Dong He distortion minimum self-responsive sound source vector The second target is the person who speaks to the ft; the driver sound source codebook, who uses M to generate the driver sound source vector corresponding to the vector length of the continuation parameter; the driver sound source retrieval mechanism • uses M to evaluate and obtain from the aforementioned driver sound source vector The relative distortion of the second target speech vector of the second speech synthesis vector, and the retrieval loss becomes the smallest driving sound source vector; and the second frame sound source generates a broadcast structure, which is the driving sound from the distortion to the smallest sea direction In ft, K is used to generate the tone sea signal of the second frame length. 3. A voice code device, which is a voice code device that divides the voice input into spectrum envelope information and sound source signal information, and uses the frame position code to code the source signal sound signal, and its characteristics are: Qiaoyou: Muji voice generation and broadcasting structure. It is a vocabulary that generates vector lengths corresponding to the length of the parameters from the Jue sound key. The paper size is based on the Chinese National Pickup Car (CNS) A4 specification (210X297mm) -1 -(Please read "note f of the back and then Ϊ copy I")-install. Book color · Six 'application for the patent model A8 B8 C8 D8 It is too late to describe the number to the evaluation target K, and it is not necessary to participate in the system. Ji Cheng as a searcher for synthesizing the K-sound of the glycoside source language ... It was driven by the author Dong Yiliang to the source of the source sound sound _ DYXII; the former degree is from the longest to the 1 The estimated sound of Xiaoya is the most driven. The most distrustful of Xiaozheng, the successor, is the discontinuity between the pre-sound and the presupposition of the source, the long inspection frame and the machine diagram. ., The source speaker's voice is the most volatile to the leg, and the period is determined to be profitable from the first section of the garden _ The special voice request should be applied as the trend 4. The coefficient is set to 0. The item of the voice is included in the logic. The number theory of the speakers who can code a sound is 0. The norm of any degree of the degree of fear of the Chinese system is _ the norm and the reference of the parameter. Please extend the application to the 5th. · The number of quantifiers for each long tone of the Burma is one of the most important terms, and every 1 continuation should be equal to the addition of the M phase S3 system '1 * * The first frame into the garden II II Fan Chengyin Lisheng dialect special request frame Shen You Rumu cut &lt; 〇 in the middle of the degree to the Diyinyuan language, Fan Youli, Mie Chengcheng, shengsheng Shen M, such as 7. The loser installs the 00 tone language item 1 any middle item 3 to Qiao Yiting should be equal Separate the degree of addition and cut, and increase the amount of input. The K degree of the tone 1 language is long, double. The life of the integral machine The long language is the most worthy goal in the middle of the number # (please first pay attention to the $ item of the wllt face and then buy it in the Qin manuscript) Μ $ The price of the goods is consumed by the freightman, including the printing of the oxygen printing device 018 by the society. What is your success?生 Ϊ Μ 人 第 _ The special degree of the face tone fan box please ask Chang ft for 8. 8. To one plus each phase of the degree is long # The most logical Xiang Yan should have a decision for each pair of K related, press ft to excite • The mechanism of the special structure is formed _ in the length of the head of the vocal sounds and words. One of the Chen items of the coder 钃 攉 音 plus 0 1 Ren Xiangzhongyin linguistics The first adult Fan Yili to enter into the _ predicate Shenyurujun 9. Ping to 11 Extend the response to make the cover length hanging To create a sound-filling frame with 1% per cent of the machine _ used in it's 29 X 10 317631 A8 B8 C8 D8 々, patented scope of the late parameter vector length of the sound source signal periodization, κ to generate a frame length sound source signal By. (Please read the note $ item on the back first and then fill in this page) 10. For the speech coding device of any one of items 1 to 3 of the patent application, where the frame sound source is generated, it is located between the frames Interpolate the sound source signal corresponding to the vector length of the delay parameter to generate the sound source signal. 11. For example, in the application of the patent code No. 1 of the speech coding device "In which the adaptive sound source retrieval mechanism is equipped with a synthesis filter * and uses the impulse response of the synthesis filter, K is repeatedly calculated and transferred from the aforementioned adaptive sound source to Dong Zhongsuo The aforementioned target speech of the obtained speech synthesis vector is relatively distorted to Dong. Printed by the Industrial and Consumer Cooperatives of the Central Bureau of Economic Affairs of the Ministry of Economic Affairs 12. A voice Zen code decoding device, which divides the voice input into spectrum envelope information and sound source signal information, and encodes the sound source signal information in iiO frame units and then decodes The coded sound source signal information K generates a voice encoding and decoding device for round-robin speech, which is characterized in that the encoding side is provided with a target voice generating mechanism that generates a target voice from the voice input corresponding to the vector length of the delay parameter to the hanger ; Adaptive audio source codebook, which generates self-responsive sound source vectors corresponding to the remaining lengths of the aforementioned delay parameters from the sound source signals generated in the past; adaptive sound source retrieval mechanism, which evaluates and uses the aforementioned adaptive sound source vectors The obtained speech synthesis is relatively distorted to the aforementioned target speech vector V of Μ, and retrieves the adaptive sound source vector with the smallest distortion; and the frame sound source generating mechanism pseudo-generates the frame from the aforementioned adaptive sound source vector with the minimum distortion For the source signal of a long length, on the other hand, on the decoding side, there is a self-adaptive audio source codebook. Adaptive excitation vector by a delay to correspond to the length parameter; and frame excitation generation means, audio fold line number from those from the sound source to be rainstorm Bu Μ generated by frame length. This paper scale is suitable for China National Standard (CNS & Α4 specifications (210X297 mm) -3 _ Λ It is printed by the Ministry of Economic Affairs, Central Bureau of Economic and Trade, Beigong Consumer Cooperative Society It A8 B8 C8 D8 VI. Patent application scope 413 · δη 申 || Item 12 of the patent profile is a voice editing device, which is further equipped on the coding side, and the second object is to generate a voice structure, which is generated from the stomach # 1¾ most and from the sound source vector. The target speech is fed into the UB codebook, which is used to generate the vector length of the delay parameter «Zhi« Guan I sound source is the most: driving the sound source retrieval structure, which is obtained by using κ evaluation and ¢ # $ 18111 sound source The aforementioned = relative distortion of the second speech synthesis vector, and the search distortion becomes the smallest drive; and the second ring frame sound source generation structure is used to generate the second sigmoid frame length from the aforementioned distortion into the dynamic sound vector On the other hand, on the decoding side, there is a sound source codebook, which uses a W to generate a vector of driving sound source corresponding to the vector length of the continuation parameter; and a second wake-up frame sound source generating mechanism, which starts from The sound is poured into the light to generate the second _ The length of the sound source signal. 14. A voice code decoding device, which divides the voice key into spectrum envelope information and sound source signal information, and decodes the decoded code by round frame unit piece of sound source information information The sound sea signal information M is a device for generating voice output codes for voice output. The special feature is that it is provided on the side of the knitting side. The structure for generating the sound of the humiliating sound is to generate the vector length of the delay parameter from the voice input. The person who looks at the voice to ft; the driving sound is thin, and it is used to generate the meta-sound source of the vector length corresponding to the embedded parameter; the meta-sound source picks up the support structure, which is used to evaluate and drive by the aforementioned Tone (the relative distortion of the aforementioned speech vector of the speech synthesis vector obtained from t, and the search distortion becomes the minimum of the source vector of the moving sound; and the frame frame sound source generation structure is the moving from the aforementioned distortion to the minimum The sound source is the most used to generate the frame length of the sound source signal. The paper standard is to use the Chinese In house car (CNS &gt; A4 specification (210X297mm) -4 i ^ in-»l ^ i amtf l ^ i— HI In mi. ^^ 1 «1 ^ 1 tii ϋ · Β &gt; 1 IV ϋ ^ · (please read the note $ item on the back first and then fill in this page) 317631 A8 B8 C8 D8 6. Those who apply for patent scope, on the other hand, they have on the decoding side, and drive the audio source codebook. It is used to generate the spoofed sound source directed to δ corresponding to the vector length of the delay parameter; and the length of the ring frame is used to drive the slave system with a middle volume to the Meiyin. The machine source generates the source sound. ϋιϋι ^ Γιιί— ^ I installed II (please read the note $ item on the back and then fill in this page) to order. 丨 (I print 5 copies of this paper standard by the Chinese Ministry of Economic Affairs, Central Bureau of Economic Cooperation, Beigong Consumer Cooperative. Rate (CNS) Α4 specification (210X297mm)